Glossary
SHAP (SHapley Additive exPlanations) values are a technique for explaining individual machine-learning predictions by quantifying how much each input feature contributed to the model’s output, showing, for example, exactly which signals pushed a fraud score up or down. They’re one of the most widely used methods for making ML models explainable, which matters greatly in fraud detection and other regulated, high-stakes uses.
SHAP values are grounded in cooperative game theory (Shapley values), which fairly distributes a "payout" among contributors. Applied to ML, they fairly attribute a prediction among the input features that produced it.
For a given prediction, SHAP computes a value for each input feature representing its contribution to the outcome relative to a baseline. Positive SHAP values pushed the prediction one way (e.g., toward "fraud"), negative values the other. Summed, they account for the model’s output. The result is a clear, per-decision breakdown: this transaction scored high-risk because of an unrecognized device (large positive contribution), an unusual amount (moderate positive), and despite a trusted network (small negative). That feature-level attribution is exactly what an analyst or auditor needs.
Fraud models must be explainable (for analyst investigation, regulatory compliance, and trust) and SHAP is a leading way to achieve it. Instead of an opaque risk score, SHAP tells the analyst which signals drove it, so they can investigate efficiently and justify the decision. It also helps detect model problems: if a feature is contributing in ways that don’t make sense, or in ways that suggest bias, SHAP surfaces it. This is why platforms addressing the "black box problem" in fraud AI often use SHAP values to deliver transparency alongside accuracy.
SHAP is a specific, popular technique within the broader field of explainable AI (XAI). It’s valued for being model-agnostic (it can explain many types of models) and theoretically grounded (its fairness properties are well-defined). For teams deploying ML in fraud and identity, SHAP is a practical tool to meet the explainability that high-stakes, regulated decisions demand: turning powerful but opaque models into ones whose decisions can be understood, trusted, and defended.
What are SHAP values?
A technique that explains an ML prediction by quantifying how much each input feature contributed to the output.
How do SHAP values help fraud detection?
They show which signals drove a risk score, letting analysts investigate and justify decisions and satisfying regulatory explainability needs.
What does SHAP stand for?
SHapley Additive exPlanations, based on Shapley values from cooperative game theory.
Is SHAP the same as explainable AI?
SHAP is a specific, widely-used technique within the broader field of explainable AI (XAI).
Related: Explainable AI (XAI) · Machine Learning for Fraud · AI in Fraud Prevention · Risk Scoring
